Referee's scores

Referee's scores in output files will range from -2 to 91.

Since Referee's score is a ratio of probabilities, theoretically any score from -infinity to +infinity is possible. However, in practice scores tend to be limited to -300 to +300. Positive scores indicate support for the called reference base while negative scores indicate more support for some other base.

We cap Referee's scores from 0 to 90. This means that all sites that score negative are converted to a score of 0, and any sites that score higher than 90 simply get a score of 90. This scaling is practical because it makes scores easily interpratable and allows for the converstion to ASCII characters in the FASTQ output.

However, here are several scenarios where the scoring calculation is impossible. We have reserved some scores for these situations.

The sum of the genotype likelihoods that do not contain the reference base is 0 91
The reference base is called as N -1
No reads mapped to this site -2

For the case of the reference base being called as N, Referee can calculate the highest scoring base and report it with the --correct option.

Referee's scores are not Phred-scaled error probabilities

Rather, they are a a ratio of probabilities. While Referee's scores are said to be Phred-like because higher scores indicate more confidence in the called base, they do not represent a probability of error as traditional Phred scores do, and as such they are scaled differently.

Put succinctly:

    A Phred score of 10 corresponds to a probability of error of 0.1.
    A Referee score of 10 means that the called base is \(10^{10}\) times more likely than another base.